Re: [PATCH v1 0/2] vfio/pci: expose device's PASID capability to VMs

From: Yan Zhao
Date: Thu Apr 16 2020 - 21:23:18 EST


On Fri, Apr 17, 2020 at 06:33:54AM +0800, Raj, Ashok wrote:
> Hi Zhao
>
>
> On Thu, Apr 16, 2020 at 06:12:26PM -0400, Yan Zhao wrote:
> > On Tue, Mar 31, 2020 at 03:08:25PM +0800, Lu, Baolu wrote:
> > > On 2020/3/31 14:35, Tian, Kevin wrote:
> > > >> From: Liu, Yi L<yi.l.liu@xxxxxxxxx>
> > > >> Sent: Sunday, March 22, 2020 8:33 PM
> > > >>
> > > >> From: Liu Yi L<yi.l.liu@xxxxxxxxx>
> > > >>
> > > >> Shared Virtual Addressing (SVA), a.k.a, Shared Virtual Memory (SVM) on
> > > >> Intel platforms allows address space sharing between device DMA and
> > > >> applications. SVA can reduce programming complexity and enhance security.
> > > >>
> > > >> To enable SVA, device needs to have PASID capability, which is a key
> > > >> capability for SVA. This patchset exposes the device's PASID capability
> > > >> to guest instead of hiding it from guest.
> > > >>
> > > >> The second patch emulates PASID capability for VFs (Virtual Function) since
> > > >> VFs don't implement such capability per PCIe spec. This patch emulates such
> > > >> capability and expose to VM if the capability is enabled in PF (Physical
> > > >> Function).
> > > >>
> > > >> However, there is an open for PASID emulation. If PF driver disables PASID
> > > >> capability at runtime, then it may be an issue. e.g. PF should not disable
> > > >> PASID capability if there is guest using this capability on any VF related
> > > >> to this PF. To solve it, may need to introduce a generic communication
> > > >> framework between vfio-pci driver and PF drivers. Please feel free to give
> > > >> your suggestions on it.
> > > > I'm not sure how this is addressed on bate metal today, i.e. between normal
> > > > kernel PF and VF drivers. I look at pasid enable/disable code in intel-iommu.c.
> > > > There is no check on PF/VF dependency so far. The cap is toggled when
> > > > attaching/detaching the PF to its domain. Let's see how IOMMU guys
> > > > respond, and if there is a way for VF driver to block PF driver from disabling
> > > > the pasid cap when it's being actively used by VF driver, then we may
> > > > leverage the same trick in VFIO when emulation is provided to guest.
> > >
> > > IOMMU subsystem doesn't expose any APIs for pasid enabling/disabling.
> > > The PCI subsystem does. It handles VF/PF like below.
> > >
> > > /**
> > > * pci_enable_pasid - Enable the PASID capability
> > > * @pdev: PCI device structure
> > > * @features: Features to enable
> > > *
> > > * Returns 0 on success, negative value on error. This function checks
> > > * whether the features are actually supported by the device and returns
> > > * an error if not.
> > > */
> > > int pci_enable_pasid(struct pci_dev *pdev, int features)
> > > {
> > > u16 control, supported;
> > > int pasid = pdev->pasid_cap;
> > >
> > > /*
> > > * VFs must not implement the PASID Capability, but if a PF
> > > * supports PASID, its VFs share the PF PASID configuration.
> > > */
> > > if (pdev->is_virtfn) {
> > > if (pci_physfn(pdev)->pasid_enabled)
> > > return 0;
> > > return -EINVAL;
> > > }
> > >
> > > /**
> > > * pci_disable_pasid - Disable the PASID capability
> > > * @pdev: PCI device structure
> > > */
> > > void pci_disable_pasid(struct pci_dev *pdev)
> > > {
> > > u16 control = 0;
> > > int pasid = pdev->pasid_cap;
> > >
> > > /* VFs share the PF PASID configuration */
> > > if (pdev->is_virtfn)
> > > return;
> > >
> > >
> > > It doesn't block disabling PASID on PF even VFs are possibly using it.
> > >
> > hi
> > I'm not sure, but is it possible for pci_enable_pasid() and
> > pci_disable_pasid() to do the same thing as pdev->driver->sriov_configure,
> > e.g. pci_sriov_configure_simple() below.
> >
> > It checks whether there are VFs are assigned in pci_vfs_assigned(dev).
> > and we can set the VF in assigned status if vfio_pci_open() is performed
> > on the VF.
>
> But you can still unbind the PF driver that magically causes the VF's to be
> removed from the guest image too correct?
>
> Only the IOMMU mucks with pasid_enable/disable. And it doesn't look like
> we have a path to disable without tearing down the PF binding.
>
> We originally had some refcounts and such and would do the real disable only
> when the refcount drops to 0, but we found it wasn't actually necessary
> to protect these resources like that.
>
right. now unbinding PF driver would cause VFs unplugged from guest.
if we modify vfio_pci and set VFs to be assigned, then VFs could remain
appearing in guest but it cannot function well as PF driver has been unbound.

thanks for explanation :)

> >
> >
> > int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
> > {
> > int rc;
> >
> > might_sleep();
> >
> > if (!dev->is_physfn)
> > return -ENODEV;
> >
> > if (pci_vfs_assigned(dev)) {
> > pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
> > return -EPERM;
> > }
> >
> > if (nr_virtfn == 0) {
> > sriov_disable(dev);
> > return 0;
> > }
> >
> > rc = sriov_enable(dev, nr_virtfn);
> > if (rc < 0)
> > return rc;
> >
> > return nr_virtfn;
> > }
> >
> > Thanks
> > Yan